A Neural Model for Contextual Biasing Score Learning and Filtering

2025-10-27

AI Summary

This work addresses the challenge of effectively integrating external knowledge—such as user-specific phrases or entities—into the decoding process of personalized automatic speech recognition (ASR). We propose an attention-bias-based candidate phrase filtering framework that dynamically scores candidate phrase tokens using ASR encoder outputs, and jointly leverages shallow fusion with a discriminative token-level loss to precisely suppress low-confidence distractors. The framework is modular and plug-and-play, compatible with any encoder-decoder ASR architecture. Evaluated on the LibriSpeech biasing benchmark, it achieves over 90% candidate redundancy reduction while substantially improving word error rate (WER) across diverse biasing strengths. Our approach establishes a general, efficient, and interpretable paradigm for knowledge-guided ASR biasing, offering both robust performance gains and transparent control over external knowledge integration.

Technology Category

• Natural Language Processing: Language Grounding & Multi-modal NLP
• Machine Learning: Learning Preferences or Rankings
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Contextual biasing improves automatic speech recognition (ASR) by integrating external knowledge, such as user-specific phrases or entities, during decoding. In this work, we use an attention-based biasing decoder to produce scores for candidate phrases based on acoustic information extracted by an ASR encoder, which can be used to filter out unlikely phrases and to calculate bonus for shallow-fusion biasing. We introduce a per-token discriminative objective that encourages higher scores for ground-truth phrases while suppressing distractors. Experiments on the Librispeech biasing benchmark show that our method effectively filters out majority of the candidate phrases, and significantly improves recognition accuracy under different biasing conditions when the scores are used in shallow fusion biasing. Our approach is modular and can be used with any ASR system, and the filtering mechanism can potentially boost performance of other biasing methods.

Problem

• Improves speech recognition by integrating external knowledge during decoding
• Filters unlikely phrases using attention-based scores from acoustic information
• Enhances recognition accuracy across different biasing conditions modularly

Innovation

• Attention-based decoder scores candidate phrases acoustically
• Per-token discriminative objective suppresses distractors
• Modular filtering mechanism works with any ASR system